The Rule, Not the Reading

On Monday, Common Sense Media released the first Census of AI use by American tweens and teens.¹ The fieldwork was done in winter and spring. The report landed from the organization's newly created Youth AI Safety Institute, the independent testing lab the group launched in May to do what no one else had bothered to do yet, look closely at the products the field had already shipped to children.²

The headline number is eighty-six percent. Eighty-six percent of kids ages nine to seventeen have used a generative AI tool. Eighty-one percent of the nine to twelve group. Eighty-nine percent of the thirteen to fifteen group. Ninety-two percent of the sixteen to seventeen group. Nearly a quarter use AI every day.¹ The number is meant to land like a bell. It probably will.

The number worth dwelling on is two pages deeper. Three-quarters of kids say their school has told them what they can and cannot use AI for. Only about half say their school has taught them how to tell whether what an AI tool says is true. Only a third knew that the systems they have been using cannot reliably distinguish fact from fiction.¹

That gap is the story. The rule was taught. The reading was not.

The Year That Skipped a Step

The field has had three years to write AI policies, and most districts now have one. By recent counts the share of districts without written AI guidance has dropped from forty-three percent last year to twenty-one percent this year. Ohio's House Bill 96 will require every public, community, and STEM school in the state to adopt a formal AI policy by July 1, the first such state mandate in the country.³ The arrival of the rule is the visible sign of seriousness.

The Census data is a sober reminder of what the rule alone does. It tells a child what is forbidden. It does not tell her how to read what the tool gives her. Half of the children answering the survey know they are not supposed to copy from ChatGPT for a science worksheet. Only a third know that the same ChatGPT, used for the same worksheet, will sometimes invent the citation, the chemical, the date.¹

That second skill is the one that matters more, because the first one is enforceable only when adults are looking. The second one travels with the student into the evening, into the bedroom, into the half of her day no rubric reaches.

The Asymmetry of Confidence

There is a smaller, sadder data point inside the Census. More than half of kids surveyed say they turn to AI for advice on their health or body.¹ They do this in the absence of a developed understanding that the tool is, on a clinical level, not a reliable narrator. The school did the harder part of the work, drafting the policy. It did not yet do the smaller, more useful part, which is the slow building of a reader.

This is what it looks like when the conversation about AI in schools gets organized around what is allowed before it is organized around what is true. The policy answers a question students were not, on the whole, asking. The discernment answers the question they ask the tool a dozen times a day, often without knowing they are asking it.

A Future That Reads What It Uses

The classrooms that will get through the next ten years are the ones that treat discernment as a teachable skill, the way grammar and proof and source citation were treated for the previous hundred years. Not a single slide on hallucination at the start of the unit. A daily practice. The student writes a paragraph with the help of the tool. She marks which sentence she sourced from the model. She checks the source the model gave her. She notes whether it held up. The teacher sees the trail of that checking, and the next prompt asks her to do it again.

This is the part of the answer Koan was quietly built for. Not the rule about which assignments allow AI. The record of how a student read the AI when she used it. The pause before she accepted the model's first sentence. The revision after she found the model's citation did not exist. The note she made to herself when she caught it. None of that is visible in a finished essay. All of it is the skill the Census tells us has not yet been taught.

The eighty-six percent are not the warning. They are the inventory. The warning is the half who were given the rule and not the reading.

If a child in your building used AI on three assignments this week, would you be able to say which of them she read carefully and which she did not?